appConfigEntityId

bodyType

customContents

entityId

entityUuid

isVideo

articlePageQuery

article

articleSchemaContent

content

hooksArticlePageAdvertisementConfigArticle

hooksArticlePageFloatingButtonConfigArticle

appConfig

filter

json

liveArticle

Xinmei Shen

A flagship model from China’s Zhipu AI has helped contain an autonomous cyberattack by OpenAI’s frontier systems targeting popular developer platform Hugging Face, as concerns grow over the security risks posed by advanced AI models.
OpenAI’s latest flagship models – including GPT-5.6 Sol and an unreleased, “even more capable” system – recently breached Hugging Face’s infrastructure during internal evaluations of their offensive cyber capabilities, the US lab disclosed on Wednesday.
The company said its models were operating in a sandboxed environment – an isolated virtual testing ground – designed to solve challenges from ExploitGym, a leading cybersecurity benchmark developed by researchers at the University of California, Berkeley, led by renowned Chinese-American scientist Dawn Song.
Upon inferring that Hugging Face hosted potential solutions to the benchmark tests, the models “successfully found ways to gain access to secret information that [they] could use to cheat the evaluation”, OpenAI said. It described the event as an “unprecedented cyber incident”.
Hugging Face, the New York-headquartered platform widely used for open-source AI collaboration, first disclosed the breach last week without naming the source.
The intrusion was “different from anything we had handled before in one important way: it was driven, end-to-end, by an autonomous AI agent system”, the company said in a blog post last Thursday.

When Hugging Face detected the intrusion, its initial defensive tactic was to deploy frontier proprietary models via commercial application programming interfaces (APIs), but those requests were blocked by the providers’ automated safety guardrails, it said.
Consequently, the platform opted to run Zhipu’s open-weight model GLM-5.2 on its own internal hardware to contain the attack.
OpenAI said on Wednesday that Hugging Face had already detected and halted the breach by the time the two firms established contact regarding the incident, adding that a joint investigation was ongoing.
Hugging Face co-founder and CEO Clement Delangue wrote on X on Wednesday that the company “strongly believes there was no malicious intent” on OpenAI’s part.
Still, the incident has fuelled growing debate over the rapid advancement of autonomous AI systems capable of exploiting software vulnerabilities, while highlighting both the security risks and benefits of open-weight models.
It also comes as the US has tightened its approach towards managing frontier AI risks.
OpenAI rival Anthropic, for example, restricted access to its top-tier Mythos model to select US partners while automatically downgrading sensitive tasks – including cybersecurity requests – on its Fable 5 model to an older system.
While safety researchers remain concerned that frontier open-weight models could lower technical barriers for malicious actors, the incident has spurred calls advocating for open-source tools in network defence.
“One takeaway, we fought back with open models, in the open,” Hugging Face head of infrastructure Adrien Carreira wrote on X on Wednesday. “AI security won’t be solved by one company in secret. Open source puts these tools in every defender’s hands.”
Hugging Face co-founder and chief science officer Thomas Wolf echoed the sentiment, noting that the event “reinforced my belief in the importance of access to capable open-weight models for cyber defence”, describing open-source AI as one of the “strongest tools” for securing the ecosystem.
However, Clement Neo, an AI safety researcher and founder of Neo Research, noted that the breach underscored both the benefits and risks inherent to open-source models, as well as how “unprepared everyone is”.
The industry was not ready “both in terms of how closed-source models are gated today, and how open-source models are useful today but potentially dangerous in the near future”, Neo said.


Philippine tribe vows to stand ground against US-led Pax Silica: ‘listen to us’

As the US weighs restrictions on cloud computing, how will China’s AI sector adapt?

Chinese AI chipmaker Hygon plots expansion from data centres to robotics

For China’s hi-tech IPOs, it’s Hong Kong vs the mainland – but how much does it matter?

Artificial intelligence (AI)

Hugging Face deploys Zhipu’s GLM 5.2 model to contain autonomous OpenAI cyberattack

The OpenAI logo is seen in this illustration taken June 11, 2026. Photo: Reuters

The Zhipu AI logo seen at its headquarters during a government-organised media tour in Beijing, June 25, 2026. Photo: Reuters

China’s Kimi K3 fuels fears safety curbs are holding back US AI

People walk past the Moonshot booth promoting its Kimi K3 AI model during the World Artificial Intelligence Conference in Shanghai on July 17. Photo: Reuters

As US-Chinese AI model gap narrows, what next for Washington?

China’s latest AI model advancements have forced Washington to ponder its next move in the global artificial intelligence race. Image: Shutterstock

Tech

Tech Trends

News

China Future Tech

The incident fuels growing debate over the rapid advancement of autonomous AI systems capable of exploiting software vulnerabilities


The incident fuels growing debate over the rapid advancement of autonomous AI systems capable of exploiting software vulnerabilities.


Hugging Face deploys Zhipu’s GLM 5.2 model to contain autonomous OpenAI cyberattack

.css-1c6uqr6{color:inherit;font-weight:inherit;font-size:inherit;font-family:inherit;line-height:inherit;overflow-wrap:break-word;}The incident fuels growing debate over the rapid advancement of autonomous AI systems capable of exploiting software vulnerabilities

The incident fuels growing debate over the rapid advancement of autonomous AI systems capable of exploiting software vulnerabilities